diff --git a/indra/aistatemachine/aistatemachine.cpp b/indra/aistatemachine/aistatemachine.cpp
index fe7b4a3bd..134461e09 100644
--- a/indra/aistatemachine/aistatemachine.cpp
+++ b/indra/aistatemachine/aistatemachine.cpp
@@ -440,7 +440,7 @@ void AIStateMachine::multiplex(event_type event)
 	// our need to run (by us having set need_run), so there is no need to run
 	// ourselves.
 	llassert(!mMultiplexMutex.isSelfLocked());		// We may never enter recursively!
-	if (!mMultiplexMutex.tryLock())
+	if (!mMultiplexMutex.try_lock())
 	{
 	  Dout(dc::statemachine(mSMDebug), "Leaving because it is already being run [" << (void*)this << "]");
 	  return;
@@ -762,7 +762,7 @@ void AIStateMachine::multiplex(event_type event)
 		//=========================================
 
 		// Release the lock on mMultiplexMutex *first*, before releasing the lock on mState,
-		// to avoid to ever call the tryLock() and fail, while this thread isn't still
+		// to avoid to ever call the try_lock() and fail, while this thread isn't still
 		// BEFORE the critical area of mState!
 
 		mMultiplexMutex.unlock();
@@ -1262,7 +1262,7 @@ void AIStateMachine::abort(void)
 	multiplex(insert_abort);
   }
   // Block until the current run finished.
-  if (!mRunMutex.tryLock())
+  if (!mRunMutex.try_lock())
   {
 	llwarns << "AIStateMachine::abort() blocks because the statemachine is still executing code in another thread." << llendl;
 	mRunMutex.lock();
diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt
index 79674c5a0..ccc48860d 100644
--- a/indra/llcommon/CMakeLists.txt
+++ b/indra/llcommon/CMakeLists.txt
@@ -284,6 +284,8 @@ target_link_libraries(
     ${WINDOWS_LIBRARIES}
     ${Boost_CONTEXT_LIBRARY}
     ${Boost_REGEX_LIBRARY}
+    ${Boost_THREAD_LIBRARY}
+    ${Boost_SYSTEM_LIBRARY}
     ${CORESERVICES_LIBRARY}
     )
 
diff --git a/indra/llcommon/aiframetimer.cpp b/indra/llcommon/aiframetimer.cpp
index 3c102a9c7..4e5c104c1 100644
--- a/indra/llcommon/aiframetimer.cpp
+++ b/indra/llcommon/aiframetimer.cpp
@@ -65,7 +65,7 @@ static F64 const NEVER = 1e16;				// 317 million years.
 
 F64 AIFrameTimer::sNextExpiration;
 AIFrameTimer::timer_list_type AIFrameTimer::sTimerList;
-LLMutex AIFrameTimer::sMutex;
+LLGlobalMutex AIFrameTimer::sMutex;
 
 // Notes on thread-safety of AIRunningFrameTimer (continued from aiframetimer.h)
 //
@@ -80,11 +80,10 @@ LLMutex AIFrameTimer::sMutex;
 void AIFrameTimer::create(F64 expiration, signal_type::slot_type const& slot)
 {
 	AIRunningFrameTimer new_timer(expiration, this);
-	sMutex.lock();
+	LLMutexLock lock(sMutex);
 	llassert(mHandle.mRunningTimer == sTimerList.end());	// Create may only be called when the timer isn't already running.
 	mHandle.init(sTimerList.insert(new_timer), slot);
 	sNextExpiration = sTimerList.begin()->expiration();
-	sMutex.unlock();
 }
 
 void AIFrameTimer::cancel(void)
@@ -95,18 +94,19 @@ void AIFrameTimer::cancel(void)
 	// mHandle.mMutex lock), we start with trying to obtain
 	// it here and as such wait till the callback function
 	// returned.
-    mHandle.mMutex.lock();
+	mHandle.mMutex.lock();
 	// Next we have to grab this lock in order to stop
 	// AIFrameTimer::handleExpiration from even entering
 	// in the case we manage to get it first.
-    sMutex.lock();
-    if (mHandle.mRunningTimer != sTimerList.end())
 	{
-		sTimerList.erase(mHandle.mRunningTimer);
-		mHandle.mRunningTimer = sTimerList.end();
-		sNextExpiration = sTimerList.empty() ? NEVER : sTimerList.begin()->expiration();
+		LLMutexLock lock(sMutex);
+		if (mHandle.mRunningTimer != sTimerList.end())
+		{
+			sTimerList.erase(mHandle.mRunningTimer);
+			mHandle.mRunningTimer = sTimerList.end();
+			sNextExpiration = sTimerList.empty() ? NEVER : sTimerList.begin()->expiration();
+		}
 	}
-    sMutex.unlock();
     mHandle.mMutex.unlock();
 }
 
@@ -164,7 +164,7 @@ void AIFrameTimer::handleExpiration(F64 current_frame_time)
 		//
 		// Note that if the other thread actually obtained the sMutex then we
 		// can't be here: this is still inside the critical area of sMutex.
-		if (handle.mMutex.tryLock())			// If this fails then another thread is in the process of cancelling this timer, so do nothing.
+		if (handle.mMutex.try_lock())			// If this fails then another thread is in the process of cancelling this timer, so do nothing.
 		{
 			sMutex.unlock();
 			running_timer->do_callback();		// May not throw exceptions.
diff --git a/indra/llcommon/aiframetimer.h b/indra/llcommon/aiframetimer.h
index 366581c49..880b492ee 100644
--- a/indra/llcommon/aiframetimer.h
+++ b/indra/llcommon/aiframetimer.h
@@ -96,7 +96,7 @@ class LL_COMMON_API AIFrameTimer
 
 	typedef std::multiset<AIRunningFrameTimer> timer_list_type;
 
-	static LLMutex sMutex;						// Mutex for the two global variables below.
+	static LLGlobalMutex sMutex;				// Mutex for the two global variables below.
 	static timer_list_type sTimerList;			// List with all running timers.
 	static F64 sNextExpiration;					// Cache of smallest value in sTimerList.
 	friend class LLFrameTimer;					// Access to sNextExpiration.
diff --git a/indra/llcommon/aithreadsafe.h b/indra/llcommon/aithreadsafe.h
index bb0ac7e05..93b38246f 100644
--- a/indra/llcommon/aithreadsafe.h
+++ b/indra/llcommon/aithreadsafe.h
@@ -239,7 +239,7 @@ protected:
 
 	// For use by AIThreadSafeDC
 	AIThreadSafe(void) { }
-	AIThreadSafe(LLAPRPool& parent) : mRWLock(parent) { }
+	MUTEX_POOL(AIThreadSafe(LLAPRPool& parent) : mRWLock(parent){ })
 
 public:
 	// Only for use by AITHREADSAFE, see below.
@@ -473,7 +473,7 @@ protected:
 	friend struct AIRegisteredStateMachinesList;
 	// For use by AIThreadSafeSimpleDC and AIRegisteredStateMachinesList.
 	AIThreadSafeSimple(void) { }
-	AIThreadSafeSimple(LLAPRPool& parent) : mMutex(parent) { }
+	MUTEX_POOL(AIThreadSafeSimple(LLAPRPool& parent) : mMutex(parent) { })
 
 public:
 	// Only for use by AITHREADSAFESIMPLE, see below.
@@ -551,7 +551,7 @@ public:
 
 protected:
 	// For use by AIThreadSafeSimpleDCRootPool
-	AIThreadSafeSimpleDC(LLAPRRootPool& parent) : AIThreadSafeSimple<T, MUTEX>(parent) { new (AIThreadSafeSimple<T, MUTEX>::ptr()) T; }
+	AIThreadSafeSimpleDC(LLAPRRootPool& parent) : AIThreadSafeSimple<T, MUTEX>(MUTEX_POOL(parent)) { new (AIThreadSafeSimple<T, MUTEX>::ptr()) T; }
 };
 
 // Helper class for AIThreadSafeSimpleDCRootPool to assure initialization of
@@ -585,7 +585,7 @@ public:
 	// as opposed to allocated from the current threads root pool.
 	AIThreadSafeSimpleDCRootPool(void) :
 		AIThreadSafeSimpleDCRootPool_pbase(),
-		AIThreadSafeSimpleDC<T>(mRootPool) { }
+		AIThreadSafeSimpleDC<T>(MUTEX_POOL(mRootPool)) { }
 };
 
 /**
diff --git a/indra/llcommon/llapr.cpp b/indra/llcommon/llapr.cpp
index 4b86fecdf..28e8db756 100644
--- a/indra/llcommon/llapr.cpp
+++ b/indra/llcommon/llapr.cpp
@@ -30,60 +30,6 @@
 #include "llapr.h"
 #include "llscopedvolatileaprpool.h"
 
-LLFastTimer::DeclareTimer FT_WAIT_FOR_SCOPEDLOCK("LLScopedLock");
-
-//---------------------------------------------------------------------
-//
-// LLScopedLock
-//
-LLScopedLock::LLScopedLock(apr_thread_mutex_t* mutex) : mMutex(mutex)
-{
-	mLocked = !!mutex;
-	if (LL_LIKELY(mutex))
-	{
-		apr_status_t status = apr_thread_mutex_trylock(mMutex);
-		while (LL_UNLIKELY(status != APR_SUCCESS))
-		{
-			if (APR_STATUS_IS_EBUSY(status))
-			{
-				if (AIThreadID::in_main_thread_inline())
-				{
-					LLFastTimer ft1(FT_WAIT_FOR_SCOPEDLOCK);
-					status = apr_thread_mutex_lock(mMutex);
-				}
-				else
-				{
-					status = apr_thread_mutex_lock(mMutex);
-				}
-			}
-			else
-			{
-				ll_apr_warn_status(status);
-				mLocked = false;
-				return;
-			}
-		}
-	}
-}
-
-LLScopedLock::~LLScopedLock()
-{
-	unlock();
-}
-
-void LLScopedLock::unlock()
-{
-	if(mLocked)
-	{
-		if(!ll_apr_warn_status(apr_thread_mutex_unlock(mMutex)))
-		{
-			mLocked = false;
-		}
-	}
-}
-
-//---------------------------------------------------------------------
-
 bool ll_apr_warn_status(apr_status_t status)
 {
 	if(APR_SUCCESS == status) return false;
diff --git a/indra/llcommon/llapr.h b/indra/llcommon/llapr.h
index 97aa96e0b..9baddef5f 100644
--- a/indra/llcommon/llapr.h
+++ b/indra/llcommon/llapr.h
@@ -36,55 +36,12 @@
 #include <boost/noncopyable.hpp>
 
 #include "apr_thread_proc.h"
-#include "apr_thread_mutex.h"
 #include "apr_getopt.h"
-#include "apr_signal.h"
-#include "apr_atomic.h"
 #include "llstring.h"
 
 class LLAPRPool;
 class LLVolatileAPRPool;
 
-/** 
- * @class LLScopedLock
- * @brief Small class to help lock and unlock mutexes.
- *
- * This class is used to have a stack level lock once you already have
- * an apr mutex handy. The constructor handles the lock, and the
- * destructor handles the unlock. Instances of this class are
- * <b>not</b> thread safe.
- */
-class LL_COMMON_API LLScopedLock : private boost::noncopyable
-{
-public:
-	/**
-	 * @brief Constructor which accepts a mutex, and locks it.
-	 *
-	 * @param mutex An allocated APR mutex. If you pass in NULL,
-	 * this wrapper will not lock.
-	 */
-	LLScopedLock(apr_thread_mutex_t* mutex);
-
-	/**
-	 * @brief Destructor which unlocks the mutex if still locked.
-	 */
-	~LLScopedLock();
-
-	/** 
-	 * @brief Check lock.
-	 */
-	bool isLocked() const { return mLocked; }
-
-	/** 
-	 * @brief This method unlocks the mutex.
-	 */
-	void unlock();
-
-protected:
-	bool mLocked;
-	apr_thread_mutex_t* mMutex;
-};
-
 // File IO convenience functions.
 // Returns NULL if the file fails to open, sets *sizep to file size if not NULL
 // abbreviated flags
diff --git a/indra/llcommon/llaprpool.cpp b/indra/llcommon/llaprpool.cpp
index 3dffa8300..b065a5801 100644
--- a/indra/llcommon/llaprpool.cpp
+++ b/indra/llcommon/llaprpool.cpp
@@ -37,6 +37,7 @@
 
 #include "llerror.h"
 #include "llaprpool.h"
+#include "llatomic.h"
 #include "llthread.h"
 
 // Create a subpool from parent.
@@ -110,23 +111,18 @@ LLAPRInitialization::LLAPRInitialization(void)
 }
 
 bool LLAPRRootPool::sCountInitialized = false;
-apr_uint32_t volatile LLAPRRootPool::sCount;
-
-apr_thread_mutex_t* gLogMutexp;
-apr_thread_mutex_t* gCallStacksLogMutexp;
+LLAtomicS32 LLAPRRootPool::sCount;
 
 LLAPRRootPool::LLAPRRootPool(void) : LLAPRInitialization(), LLAPRPool(0)
 {
 	// sCountInitialized don't need locking because when we get here there is still only a single thread.
 	if (!sCountInitialized)
 	{
-		// Initialize the logging mutex
-		apr_thread_mutex_create(&gLogMutexp, APR_THREAD_MUTEX_UNNESTED, mPool);
-		apr_thread_mutex_create(&gCallStacksLogMutexp, APR_THREAD_MUTEX_UNNESTED, mPool);
-
+#ifdef NEEDS_APR_ATOMICS
 		apr_status_t status = apr_atomic_init(mPool);
 		llassert_always(status == APR_SUCCESS);
-		apr_atomic_set32(&sCount, 1);	// Set to 1 to account for the global root pool.
+#endif
+		sCount = 1;	// Set to 1 to account for the global root pool.
 		sCountInitialized = true;
 
 		// Initialize thread-local APR pool support.
@@ -134,33 +130,16 @@ LLAPRRootPool::LLAPRRootPool(void) : LLAPRInitialization(), LLAPRPool(0)
 		// it must be done last, so that sCount is already initialized.
 		LLThreadLocalData::init();
 	}
-	apr_atomic_inc32(&sCount);
+	sCount++;
 }
 
 LLAPRRootPool::~LLAPRRootPool()
 {
-	if (!apr_atomic_dec32(&sCount))
+	if (!--sCount)
 	{
 		// The last pool was destructed. Cleanup remainder of APR.
 		LL_INFOS("APR") << "Cleaning up APR" << LL_ENDL;
 
-		if (gLogMutexp)
-		{
-			// Clean up the logging mutex
-
-			// All other threads NEED to be done before we clean up APR, so this is okay.
-			apr_thread_mutex_destroy(gLogMutexp);
-			gLogMutexp = NULL;
-		}
-		if (gCallStacksLogMutexp)
-		{
-			// Clean up the logging mutex
-
-			// All other threads NEED to be done before we clean up APR, so this is okay.
-			apr_thread_mutex_destroy(gCallStacksLogMutexp);
-			gCallStacksLogMutexp = NULL;
-		}
-
 		// Must destroy ALL, and therefore this last LLAPRRootPool, before terminating APR.
 		static_cast<LLAPRRootPool*>(this)->destroy();
 
diff --git a/indra/llcommon/llaprpool.h b/indra/llcommon/llaprpool.h
index 74af351e4..d3606d396 100644
--- a/indra/llcommon/llaprpool.h
+++ b/indra/llcommon/llaprpool.h
@@ -47,9 +47,11 @@
 
 #include "apr_portable.h"
 #include "apr_pools.h"
+#include "llatomic.h"
 #include "llerror.h"
 #include "aithreadid.h"
 
+
 extern void ll_init_apr();
 
 /**
@@ -104,8 +106,11 @@ public:
 	// NEVER destroy a pool that is returned by this function!
 	apr_pool_t* operator()(void) const
 	{
-		llassert(mPool);
-		llassert(mOwner.equals_current_thread());
+		if (mParent)
+		{
+			llassert(mPool);
+			llassert(mOwner.equals_current_thread());
+		}
 		return mPool;
 	}
 
@@ -181,7 +186,7 @@ private:
 private:
 	// Keep track of how many root pools exist and when the last one is destructed.
 	static bool sCountInitialized;
-	static apr_uint32_t volatile sCount;
+	static LLAtomicS32 sCount;
 
 public:
 	// Return a global root pool that is independent of LLThreadLocalData.
diff --git a/indra/llcommon/llatomic.h b/indra/llcommon/llatomic.h
index 3cc5bcf38..0cfe99c77 100644
--- a/indra/llcommon/llatomic.h
+++ b/indra/llcommon/llatomic.h
@@ -34,8 +34,29 @@
 #ifndef LL_LLATOMIC_H
 #define LL_LLATOMIC_H
 
-#include "apr_atomic.h"
+#define USE_BOOST_ATOMIC
 
+//Internal definitions
+#define NEEDS_APR_ATOMICS do_not_define_manually_thanks
+#undef NEEDS_APR_ATOMICS
+
+#if defined(USE_BOOST_ATOMIC)
+#include "boost/version.hpp"
+#endif
+
+//Prefer boost over stl over apr.
+
+#if defined(USE_BOOST_ATOMIC) && (BOOST_VERSION >= 105200)
+#include "boost/atomic.hpp"
+template<typename T>
+struct impl_atomic_type { typedef boost::atomic<T> type; };
+#elif defined(USE_STD_ATOMIC) && (__cplusplus >= 201103L || _MSC_VER >= 1800)
+#include <atomic>
+template<typename T>
+struct impl_atomic_type { typedef std::atomic<T> type; };
+#else
+#include "apr_atomic.h"
+#define NEEDS_APR_ATOMICS
 template <typename Type> class LLAtomic32
 {
 public:
@@ -54,6 +75,27 @@ public:
 private:
 	apr_uint32_t mData;
 };
+#endif
+#if !defined(NEEDS_APR_ATOMICS)
+template <typename Type> class LLAtomic32
+{
+public:
+	LLAtomic32(void) { }
+	LLAtomic32(LLAtomic32 const& atom) { mData = Type(atom.mData); }
+	LLAtomic32(Type x) { mData = x; }
+	LLAtomic32& operator=(LLAtomic32 const& atom) { mData = Type(atom.mData); return *this; }
+
+	operator Type() const { return mData; }
+	void operator=(Type x) { mData = x; }
+	void operator-=(Type x) { mData -= x; }
+	void operator+=(Type x) { mData += x; }
+	Type operator++(int) { return mData++; } // Type++
+	bool operator--() { return --mData; } // Returns (--Type != 0)
+
+private:
+	typename impl_atomic_type<Type>::type mData;
+};
+#endif
 
 typedef LLAtomic32<U32> LLAtomicU32;
 typedef LLAtomic32<S32> LLAtomicS32;
diff --git a/indra/llcommon/llerror.cpp b/indra/llcommon/llerror.cpp
index 8b40edb40..4155885d8 100644
--- a/indra/llcommon/llerror.cpp
+++ b/indra/llcommon/llerror.cpp
@@ -874,8 +874,8 @@ You get:
 	
 */
 
-extern apr_thread_mutex_t* gLogMutexp;
-extern apr_thread_mutex_t* gCallStacksLogMutexp;
+LLGlobalMutex gLogMutex;
+LLGlobalMutex gCallStacksLogMutex;
 
 namespace {
 	bool checkLevelMap(const LevelMap& map, const std::string& key,
@@ -905,17 +905,15 @@ namespace {
 	LogLock::LogLock()
 		: mLocked(false), mOK(false)
 	{
-		if (!gLogMutexp)
+		if (!gLogMutex.isInitalized())
 		{
 			mOK = true;
 			return;
 		}
-		
 		const int MAX_RETRIES = 5;
 		for (int attempts = 0; attempts < MAX_RETRIES; ++attempts)
 		{
-			apr_status_t s = apr_thread_mutex_trylock(gLogMutexp);
-			if (!APR_STATUS_IS_EBUSY(s))
+			if (gLogMutex.try_lock())
 			{
 				mLocked = true;
 				mOK = true;
@@ -937,7 +935,7 @@ namespace {
 	{
 		if (mLocked)
 		{
-			apr_thread_mutex_unlock(gLogMutexp);
+			gLogMutex.unlock();
 		}
 	}
 }
@@ -1299,17 +1297,16 @@ namespace LLError
 	CallStacksLogLock::CallStacksLogLock()
 		: mLocked(false), mOK(false)
 	{
-		if (!gCallStacksLogMutexp)
+		if (!gCallStacksLogMutex.isInitalized())
 		{
 			mOK = true;
 			return;
 		}
-		
+
 		const int MAX_RETRIES = 5;
 		for (int attempts = 0; attempts < MAX_RETRIES; ++attempts)
 		{
-			apr_status_t s = apr_thread_mutex_trylock(gCallStacksLogMutexp);
-			if (!APR_STATUS_IS_EBUSY(s))
+			if (gCallStacksLogMutex.try_lock())
 			{
 				mLocked = true;
 				mOK = true;
@@ -1328,7 +1325,7 @@ namespace LLError
 	{
 		if (mLocked)
 		{
-			apr_thread_mutex_unlock(gCallStacksLogMutexp);
+			gCallStacksLogMutex.unlock();
 		}
 	}
 
diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp
index c214f03a4..94f0c1dee 100644
--- a/indra/llcommon/llframetimer.cpp
+++ b/indra/llcommon/llframetimer.cpp
@@ -47,12 +47,14 @@ S32 LLFrameTimer::sFrameCount = 0;									// Current frame number (number of fr
 U64 LLFrameTimer::sPrevTotalTime = LLFrameTimer::sStartTotalTime;	// Previous (frame) time in microseconds since epoch, updated once per frame.
 U64 LLFrameTimer::sFrameDeltaTime = 0;								// Microseconds between last two calls to LLFrameTimer::updateFrameTimeAndCount.
 // Mutex for the above.
-apr_thread_mutex_t* LLFrameTimer::sGlobalMutex;
+LLGlobalMutex LLFrameTimer::sGlobalMutex;
+
+bool LLFrameTimer::sFirstFrameTimerCreated;
 
 // static
 void LLFrameTimer::global_initialization(void)
 {
-	apr_thread_mutex_create(&sGlobalMutex, APR_THREAD_MUTEX_UNNESTED, LLAPRRootPool::get()());
+	sFirstFrameTimerCreated = true;
 	AIFrameTimer::sNextExpiration = NEVER;
 }
 
@@ -63,9 +65,9 @@ void LLFrameTimer::updateFrameTime(void)
 	sTotalTime = totalTime();
 	sTotalSeconds = U64_to_F64(sTotalTime) * USEC_TO_SEC_F64;
 	F64 new_frame_time = U64_to_F64(sTotalTime - sStartTotalTime) * USEC_TO_SEC_F64;
-	apr_thread_mutex_lock(sGlobalMutex);
+	sGlobalMutex.lock();
 	sFrameTime = new_frame_time;
-	apr_thread_mutex_unlock(sGlobalMutex);
+	sGlobalMutex.unlock();
 }
 
 // static
diff --git a/indra/llcommon/llframetimer.h b/indra/llcommon/llframetimer.h
index 30bc58dd9..5ebf14247 100644
--- a/indra/llcommon/llframetimer.h
+++ b/indra/llcommon/llframetimer.h
@@ -36,16 +36,13 @@
 
 #include "lltimer.h"
 #include "timing.h"
-#include <apr_thread_mutex.h>
-#ifdef SHOW_ASSERT
-#include "aithreadid.h"			// is_main_thread()
-#endif
+#include "llthread.h"
 
 class LL_COMMON_API LLFrameTimer
 {
 public:
 	// Create an LLFrameTimer and start it. After creation it is running and in the state expired (hasExpired will return true).
-	LLFrameTimer(void) : mExpiry(0), mRunning(true), mPaused(false) { if (!sGlobalMutex) global_initialization(); setAge(0.0); }
+	LLFrameTimer(void) : mExpiry(0), mRunning(true), mPaused(false) { if (!sFirstFrameTimerCreated) global_initialization(); setAge(0.0); }
 
 	//<singu>
 	void copy(LLFrameTimer const& timer) { mStartTime = timer.mStartTime; mExpiry = timer.mExpiry; mRunning = timer.mRunning; mPaused = timer.mPaused; }
@@ -57,9 +54,9 @@ public:
 	static F64 getElapsedSeconds(void)
 	{
 		// Loses msec precision after ~4.5 hours...
-		apr_thread_mutex_lock(sGlobalMutex);
+		sGlobalMutex.lock();
 		F64 res = sFrameTime;
-		apr_thread_mutex_unlock(sGlobalMutex);
+		sGlobalMutex.unlock();
 		return res;
 	} 
 
@@ -68,9 +65,9 @@ public:
 	{
 		// sTotalTime is only accessed by the main thread, so no locking is necessary.
 		llassert(is_main_thread());
-		//apr_thread_mutex_lock(sGlobalMutex);
+		//sGlobalMutex.lock();
 		U64 res = sTotalTime;
-		//apr_thread_mutex_unlock(sGlobalMutex);
+		//sGlobalMutex.unlock();
 		llassert(res);
 		return res;
 	}
@@ -80,9 +77,9 @@ public:
 	{
 		// sTotalSeconds is only accessed by the main thread, so no locking is necessary.
 		llassert(is_main_thread());
-		//apr_thread_mutex_lock(sGlobalMutex);
+		//sGlobalMutex.lock();
 		F64 res = sTotalSeconds;
-		//apr_thread_mutex_unlock(sGlobalMutex);
+		//sGlobalMutex.unlock();
 		return res;
 	}
 
@@ -91,9 +88,9 @@ public:
 	{
 		// sFrameCount is only accessed by the main thread, so no locking is necessary.
 		llassert(is_main_thread());
-		//apr_thread_mutex_lock(sGlobalMutex);
+		//sGlobalMutex.lock();
 		U32 res = sFrameCount;
-		//apr_thread_mutex_unlock(sGlobalMutex);
+		//sGlobalMutex.unlock();
 		return res;
 	}
 
@@ -170,7 +167,7 @@ protected:
 	//
 
 	// More than one thread are accessing (some of) these variables, therefore we need locking.
-	static apr_thread_mutex_t* sGlobalMutex;
+	static LLGlobalMutex sGlobalMutex;
 
 	// Current time in seconds since application start, updated together with sTotalTime.
 	static F64 sFrameTime;
@@ -190,6 +187,8 @@ protected:
 	// Current frame number (number of frames since application start).
 	static S32 sFrameCount;
 
+	static bool sFirstFrameTimerCreated;
+
 	//
 	// Member data
 	//
diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h
index a780a6a24..aa8efb1e0 100644
--- a/indra/llcommon/llsingleton.h
+++ b/indra/llcommon/llsingleton.h
@@ -33,10 +33,10 @@
 
 /// @brief A global registry of all singletons to prevent duplicate allocations
 /// across shared library boundaries
-class LL_COMMON_API LLSingletonRegistry {
+class LLSingletonRegistry {
 	private:
 		typedef std::map<std::string, void *> TypeMap;
-		static TypeMap * sSingletonMap;
+		static LL_COMMON_API TypeMap * sSingletonMap;
 
 		static void checkInit()
 		{
diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp
index 1c3fec608..7ab4b1f39 100644
--- a/indra/llcommon/llthread.cpp
+++ b/indra/llcommon/llthread.cpp
@@ -364,138 +364,133 @@ LLThreadLocalData& LLThreadLocalData::tldata(void)
 
 //============================================================================
 
-LLCondition::LLCondition(LLAPRPool& parent) : LLMutex(parent)
+#if defined(NEEDS_MUTEX_IMPL)
+#if defined(USE_WIN32_THREAD)
+LLMutexImpl::LLMutexImpl()
 {
-	apr_thread_cond_create(&mAPRCondp, mPool());
+	InitializeCriticalSection(&mMutexImpl);	//can throw STATUS_NO_MEMORY
+}
+LLMutexImpl::~LLMutexImpl() 
+{
+	DeleteCriticalSection(&mMutexImpl);	//nothrow
+}
+void LLMutexImpl::lock()
+{
+	EnterCriticalSection(&mMutexImpl);	//can throw EXCEPTION_POSSIBLE_DEADLOCK
+}
+void LLMutexImpl::unlock()
+{
+	LeaveCriticalSection(&mMutexImpl);	//nothrow
+}
+bool LLMutexImpl::try_lock()
+{
+	return !!TryEnterCriticalSection(&mMutexImpl);	//nothrow
+}
+LLConditionVariableImpl::LLConditionVariableImpl()
+{
+	InitializeConditionVariable(&mConditionVariableImpl);
+}
+LLConditionVariableImpl::~LLConditionVariableImpl()
+{
+	//There is no DeleteConditionVariable
+}
+void LLConditionVariableImpl::notify_one()
+{
+	WakeConditionVariable(&mConditionVariableImpl);
+}
+void LLConditionVariableImpl::notify_all()
+{
+	WakeAllConditionVariable(&mConditionVariableImpl);
+}
+void LLConditionVariableImpl::wait(LLMutex& lock)
+{
+	LLMutex::ImplAdoptMutex impl_adopted_mutex(lock);
+	SleepConditionVariableCS(&mConditionVariableImpl, &lock.native_handle(), INFINITE);
+}
+#else
+
+void APRExceptionThrower(apr_status_t status)
+{
+	if(status != APR_SUCCESS)
+	{
+		static char buf[256];
+		throw std::logic_error(apr_strerror(status,buf,sizeof(buf)));
+	}
 }
 
-
-LLCondition::~LLCondition()
+LLMutexImpl::LLMutexImpl(native_pool_type& pool) : mPool(pool), mMutexImpl(NULL)
 {
-	apr_thread_cond_destroy(mAPRCondp);
-	mAPRCondp = NULL;
+	APRExceptionThrower(apr_thread_mutex_create(&mMutexImpl, APR_THREAD_MUTEX_UNNESTED, mPool()));
+}
+LLMutexImpl::~LLMutexImpl() 
+{
+	APRExceptionThrower(apr_thread_mutex_destroy(mMutexImpl));
+	mMutexImpl = NULL;
+}
+void LLMutexImpl::lock()
+{
+	APRExceptionThrower(apr_thread_mutex_lock(mMutexImpl));
+}
+void LLMutexImpl::unlock()
+{
+	APRExceptionThrower(apr_thread_mutex_unlock(mMutexImpl));
+}
+bool LLMutexImpl::try_lock()
+{
+	apr_status_t status = apr_thread_mutex_trylock(mMutexImpl);
+	if(APR_STATUS_IS_EBUSY(status))
+		return false;
+	APRExceptionThrower(status);
+	return true;
+}
+LLConditionVariableImpl::LLConditionVariableImpl(native_pool_type& pool) : mPool(pool), mConditionVariableImpl(NULL)
+{
+	APRExceptionThrower(apr_thread_cond_create(&mConditionVariableImpl, mPool()));
+}
+LLConditionVariableImpl::~LLConditionVariableImpl()
+{
+	APRExceptionThrower(apr_thread_cond_destroy(mConditionVariableImpl));
+}
+void LLConditionVariableImpl::notify_one()
+{
+	APRExceptionThrower(apr_thread_cond_signal(mConditionVariableImpl));
+}
+void LLConditionVariableImpl::notify_all()
+{
+	APRExceptionThrower(apr_thread_cond_broadcast(mConditionVariableImpl));
+}
+void LLConditionVariableImpl::wait(LLMutex& lock)
+{
+	LLMutex::ImplAdoptMutex impl_adopted_mutex(lock);
+	APRExceptionThrower(apr_thread_cond_wait(mConditionVariableImpl, lock.native_handle()));
+}
+#endif
+#endif
+
+LLFastTimer::DeclareTimer FT_WAIT_FOR_MUTEX("LLMutex::lock()");
+void LLMutex::lock_main(LLFastTimer::DeclareTimer* timer)
+{
+	llassert(!isSelfLocked());
+	LLFastTimer ft1(timer ? *timer : FT_WAIT_FOR_MUTEX);
+	LLMutexImpl::lock();
 }
 
 LLFastTimer::DeclareTimer FT_WAIT_FOR_CONDITION("LLCondition::wait()");
-
-void LLCondition::wait()
+void LLCondition::wait_main()
 {
-	if (AIThreadID::in_main_thread_inline())
+	llassert(isSelfLocked());
+	LLFastTimer ft1(FT_WAIT_FOR_CONDITION);
+	LLConditionVariableImpl::wait(*this);
+	llassert(isSelfLocked());
+}
+
+LLFastTimer::DeclareTimer FT_WAIT_FOR_MUTEXLOCK("LLMutexLock::lock()");
+void LLMutexLock::lock()
+{
+	if (mMutex)
 	{
-		LLFastTimer ft1(FT_WAIT_FOR_CONDITION);
-		apr_thread_cond_wait(mAPRCondp, mAPRMutexp);
-	}
-	else
-	{
-		apr_thread_cond_wait(mAPRCondp, mAPRMutexp);
+		mMutex->lock(&FT_WAIT_FOR_MUTEXLOCK);
 	}
 }
 
-void LLCondition::signal()
-{
-	apr_thread_cond_signal(mAPRCondp);
-}
-
-void LLCondition::broadcast()
-{
-	apr_thread_cond_broadcast(mAPRCondp);
-}
-
-//============================================================================
-LLMutexBase::LLMutexBase() :
-	mCount(0),
-	mLockingThread(AIThreadID::sNone)
-{
-}
-
-bool LLMutexBase::isSelfLocked() const
-{
-	return mLockingThread.equals_current_thread_inline();
-}
-
-LLFastTimer::DeclareTimer FT_WAIT_FOR_MUTEX("LLMutexBase::lock()");
-
-void LLMutexBase::lock() 
-{ 
-	if (mLockingThread.equals_current_thread_inline())
-	{ //redundant lock
-		mCount++;
-		return;
-	}
-
-	if (APR_STATUS_IS_EBUSY(apr_thread_mutex_trylock(mAPRMutexp)))
-	{
-	  if (AIThreadID::in_main_thread_inline())
-	  {
-		LLFastTimer ft1(FT_WAIT_FOR_MUTEX);
-		apr_thread_mutex_lock(mAPRMutexp);
-	  }
-	  else
-	  {
-		apr_thread_mutex_lock(mAPRMutexp);
-	  }
-	}
-	
-	mLockingThread.reset_inline();
-}
-
-bool LLMutexBase::tryLock()
-{
-	if (mLockingThread.equals_current_thread_inline())
-	{ //redundant lock
-		mCount++;
-		return true;
-	}
-	bool success = !APR_STATUS_IS_EBUSY(apr_thread_mutex_trylock(mAPRMutexp));
-	if (success)
-	{
-		mLockingThread.reset_inline();
-	}
-	return success;
-}
-
-// non-blocking, but does do a lock/unlock so not free
-bool LLMutexBase::isLocked() const
-{
-	if (mLockingThread.equals_current_thread_inline())
-	  return false;		// A call to lock() won't block.
-	if (APR_STATUS_IS_EBUSY(apr_thread_mutex_trylock(mAPRMutexp)))
-	  return true;
-	apr_thread_mutex_unlock(mAPRMutexp);
-	return false;
-}
-
-void LLMutexBase::unlock()
-{
-	if (mCount > 0)
-	{ //not the root unlock
-		mCount--;
-		return;
-	}
-	mLockingThread = AIThreadID::sNone;
-
-	apr_thread_mutex_unlock(mAPRMutexp);
-}
-
-//----------------------------------------------------------------------------
-
-LLThreadSafeRefCount::LLThreadSafeRefCount() :
-	mRef(0)
-{
-}
-
-LLThreadSafeRefCount::~LLThreadSafeRefCount()
-{ 
-	if (mRef != 0)
-	{
-		llerrs << "deleting non-zero reference" << llendl;
-	}
-}
-
-//============================================================================
-
-LLResponder::~LLResponder()
-{
-}
-
-//============================================================================
+//----------------------------------------------------------------------------
\ No newline at end of file
diff --git a/indra/llcommon/llthread.h b/indra/llcommon/llthread.h
index 7d0cc593c..a3f95e6d8 100644
--- a/indra/llcommon/llthread.h
+++ b/indra/llcommon/llthread.h
@@ -27,6 +27,8 @@
 #ifndef LL_LLTHREAD_H
 #define LL_LLTHREAD_H
 
+#define USE_BOOST_MUTEX
+
 #if LL_GNUC
 // Needed for is_main_thread() when compiling with optimization (relwithdebinfo).
 // It doesn't hurt to just always specify it though.
@@ -35,10 +37,9 @@
 
 #include "llapp.h"
 #include "llapr.h"
-#include "llmemory.h"
-#include "apr_thread_cond.h"
 #include "llaprpool.h"
 #include "llatomic.h"
+#include "llmemory.h"
 #include "aithreadid.h"
 
 class LLThread;
@@ -177,126 +178,294 @@ protected:
 
 //============================================================================
 
-#define MUTEX_DEBUG (LL_DEBUG || LL_RELEASE_WITH_DEBUG_INFO)
+#define MUTEX_POOL(arg)
 
-#ifdef MUTEX_DEBUG
-// We really shouldn't be using recursive locks. Make sure of that in debug mode.
-#define MUTEX_FLAG APR_THREAD_MUTEX_UNNESTED
+//Internal definitions
+#define NEEDS_MUTEX_IMPL do_not_define_manually_thanks
+#undef NEEDS_MUTEX_IMPL
+#define NEEDS_MUTEX_RECURSION do_not_define_manually_thanks
+#undef NEEDS_MUTEX_RECURSION
+
+//Prefer boost over stl over windows over apr.
+
+#if defined(USE_BOOST_MUTEX) && (BOOST_VERSION >= 103400)	//condition_variable_any was added in boost 1.34
+//Define BOOST_SYSTEM_NO_DEPRECATED to avoid system_category() and generic_category() dependencies, as those won't be exported.
+#define BOOST_SYSTEM_NO_DEPRECATED
+#include <boost/thread/mutex.hpp>
+#include <boost/thread/recursive_mutex.hpp>
+#include <boost/thread/locks.hpp> 
+#include <boost/thread/condition_variable.hpp>
+typedef boost::recursive_mutex LLMutexImpl;
+typedef boost::condition_variable_any LLConditionVariableImpl;
+#elif defined(USE_STD_MUTEX) && (__cplusplus >= 201103L || _MSC_VER >= 1800)
+#include <mutex>
+typedef std::recursive_mutex LLMutexImpl;
+typedef std::condition_variable_any LLConditionVariableImpl;
+#elif defined(USE_WIN32_MUTEX) 
+typedef CRITICAL_SECTION impl_mutex_handle_type;
+typedef CONDITION_VARIABLE impl_cond_handle_type;
+#define NEEDS_MUTEX_IMPL
+#define NEEDS_MUTEX_RECURSION
 #else
-// Use the fastest platform-optimal lock behavior (can be recursive or non-recursive).
-#define MUTEX_FLAG APR_THREAD_MUTEX_DEFAULT
+//----APR specific------
+#include "apr_thread_cond.h"
+#include "apr_thread_mutex.h"
+#include "apr_signal.h"
+typedef LLAPRPool native_pool_type;
+typedef apr_thread_mutex_t* impl_mutex_handle_type;
+typedef apr_thread_cond_t* impl_cond_handle_type;
+#undef MUTEX_POOL
+#undef DEFAULT_POOL
+#define MUTEX_POOL(arg) arg
+#define NEEDS_MUTEX_IMPL
+#define NEEDS_MUTEX_RECURSION
+//END
 #endif
 
-class LL_COMMON_API LLMutexBase
+#ifdef NEEDS_MUTEX_IMPL
+
+//Impl classes are not meant to be accessed directly. They must be utilized by a parent classes.
+// They are designed to be 'clones' of their stl counterparts to facilitate simple drop-in 
+// replacement of underlying implementation (boost,std,apr,critical_sections,etc)
+// Members and member functions are all private.
+class LL_COMMON_API LLMutexImpl : private boost::noncopyable
 {
-public:
-	LLMutexBase() ;
-	
-	void lock();		// blocks
+	friend class LLMutex;
+	friend class LLCondition;
+	friend class LLConditionVariableImpl;
+
+	typedef impl_mutex_handle_type native_handle_type;
+
+	LLMutexImpl(MUTEX_POOL(native_pool_type& pool));
+	virtual ~LLMutexImpl();
+	void lock();
 	void unlock();
-	// Returns true if lock was obtained successfully.
-	bool tryLock();
-
-	// Returns true if a call to lock() would block (returns false if self-locked()).
-	bool isLocked() const;
-
-	// Returns true if locked by this thread.
-	bool isSelfLocked() const;
-
-protected:
-	// mAPRMutexp is initialized and uninitialized in the derived class.
-	apr_thread_mutex_t* mAPRMutexp;
-	mutable U32			mCount;
-	mutable AIThreadID	mLockingThread;
+	bool try_lock();
+	native_handle_type& native_handle()	{ return mMutexImpl; }
 
 private:
-	// Disallow copy construction and assignment.
-	LLMutexBase(LLMutexBase const&);
-	LLMutexBase& operator=(LLMutexBase const&);
+	native_handle_type mMutexImpl;
+	MUTEX_POOL(native_pool_type mPool);
 };
 
-class LL_COMMON_API LLMutex : public LLMutexBase
-{
-public:
-	LLMutex(LLAPRPool& parent = LLThread::tldata().mRootPool) : mPool(parent)
-	{
-		apr_thread_mutex_create(&mAPRMutexp, MUTEX_FLAG, mPool());
-	}
-	~LLMutex()
-	{
-		//this assertion erroneously triggers whenever an LLCondition is destroyed
-		//llassert(!isLocked()); // better not be locked!
-		apr_thread_mutex_destroy(mAPRMutexp);
-		mAPRMutexp = NULL;
-	}
-
-protected:
-	LLAPRPool mPool;
-};
-
-#if APR_HAS_THREADS
-// No need to use a root pool in this case.
-typedef LLMutex LLMutexRootPool;
-#else  // APR_HAS_THREADS
-class LL_COMMON_API LLMutexRootPool : public LLMutexBase
-{
-public:
-	LLMutexRootPool(void)
-	{
-		apr_thread_mutex_create(&mAPRMutexp, MUTEX_FLAG, mRootPool());
-	}
-	~LLMutexRootPool()
-	{
-#if APR_POOL_DEBUG
-		// It is allowed to destruct root pools from a different thread.
-		mRootPool.grab_ownership();
 #endif
-		llassert(!isLocked()); // better not be locked!
-		apr_thread_mutex_destroy(mAPRMutexp);
-		mAPRMutexp = NULL;
+
+class LL_COMMON_API LLMutex : public LLMutexImpl
+{
+#ifdef NEEDS_MUTEX_IMPL
+	friend class LLConditionVariableImpl;
+#endif
+public:
+	LLMutex(MUTEX_POOL(native_pool_type& pool = LLThread::tldata().mRootPool)) : LLMutexImpl(MUTEX_POOL(pool)),
+#ifdef NEEDS_MUTEX_RECURSION
+		mLockDepth(0),
+#endif
+		mLockingThread(AIThreadID::sNone)
+	{}
+	~LLMutex()
+	{}
+
+	void lock(LLFastTimer::DeclareTimer* timer = NULL)	// blocks
+	{
+		if (inc_lock_if_recursive())
+			return;
+		if (AIThreadID::in_main_thread_inline() && LLApp::isRunning())
+		{
+			if (!LLMutexImpl::try_lock())
+			{
+				lock_main(timer);
+			}
+		}
+		else
+		{
+			LLMutexImpl::lock();
+		}
+		mLockingThread.reset_inline();
 	}
 
-protected:
-	LLAPRRootPool mRootPool;
+	void unlock()
+	{
+#ifdef NEEDS_MUTEX_RECURSION
+		if (mLockDepth > 0)
+		{
+			--mLockDepth;
+			return;
+		}
+#endif
+		mLockingThread = AIThreadID::sNone;
+		LLMutexImpl::unlock();
+	}
+
+	// Returns true if lock was obtained successfully.
+	bool try_lock()
+	{
+		if (inc_lock_if_recursive())
+			return true;
+		if (!LLMutexImpl::try_lock())
+			return false;
+		mLockingThread.reset_inline();
+		return true;
+	}
+	
+	// Returns true if locked not by this thread
+	bool isLocked()
+	{
+		if (isSelfLocked())
+			return false;
+		if (LLMutexImpl::try_lock())
+		{
+			LLMutexImpl::unlock();
+			return false;
+		}
+		return true;
+	}
+	// Returns true if locked by this thread.
+	bool isSelfLocked() const
+	{
+		return mLockingThread.equals_current_thread_inline();
+	}
+
+#ifdef NEEDS_MUTEX_IMPL
+	//This is important for libraries that we cannot pass LLMutex into.
+	//For example, apr wait. apr wait unlocks and re-locks the thread, however
+	// it has no knowledge of LLMutex::mLockingThread and LLMutex::mLockDepth,
+	// and thus will leave those member variables set even after the wait internally releases the lock.
+	// Leaving those two variables set even when mutex has actually been unlocked via apr is BAD.
+	friend class ImplAdoptMutex;
+	class ImplAdoptMutex
+	{
+		friend class LLConditionVariableImpl;
+		ImplAdoptMutex(LLMutex& mutex) : mMutex(mutex), 
+#ifdef NEEDS_MUTEX_RECURSION
+			mLockDepth(mutex.mLockDepth),
+#endif
+			mLockingThread(mutex.mLockingThread)
+
+		{
+			mMutex.mLockingThread = AIThreadID::sNone;
+#ifdef NEEDS_MUTEX_RECURSION
+			mMutex.mLockDepth = 0;
+#endif
+		}
+		~ImplAdoptMutex()
+		{
+			mMutex.mLockingThread = mLockingThread;
+#ifdef NEEDS_MUTEX_RECURSION
+			mMutex.mLockDepth = mLockDepth;
+#endif
+		}
+		LLMutex& mMutex;
+		AIThreadID mLockingThread;
+#ifdef NEEDS_MUTEX_RECURSION
+		S32 mLockDepth;
+#endif
+	};
+#endif
+
+private:
+	void lock_main(LLFastTimer::DeclareTimer* timer);
+
+	bool inc_lock_if_recursive()
+	{
+#ifdef NEEDS_MUTEX_RECURSION
+		if (isSelfLocked())
+		{
+			mLockDepth++;
+			return true;
+		}
+#endif
+		return false;
+	}
+
+	mutable AIThreadID	mLockingThread;
+#ifdef NEEDS_MUTEX_RECURSION
+	LLAtomicS32 mLockDepth;
+#endif
 };
-#endif // APR_HAS_THREADS
+
+class LLGlobalMutex : public LLMutex
+{
+public:
+	LLGlobalMutex() : LLMutex(MUTEX_POOL(LLAPRRootPool::get())), mbInitalized(true)
+	{}
+	bool isInitalized() const
+	{
+		return mbInitalized;
+	}
+private:
+	bool mbInitalized;
+};
+
+#ifdef NEEDS_MUTEX_IMPL
+class LL_COMMON_API LLConditionVariableImpl : private boost::noncopyable
+{
+	friend class LLCondition;
+
+	typedef impl_cond_handle_type native_handle_type;
+
+	LLConditionVariableImpl(MUTEX_POOL(native_pool_type& pool));
+	virtual ~LLConditionVariableImpl();
+	void notify_one();
+	void notify_all();
+	void wait(LLMutex& lock);
+	native_handle_type& native_handle()	{ return mConditionVariableImpl; }
+
+	native_handle_type mConditionVariableImpl;
+	MUTEX_POOL(native_pool_type mPool);
+};
+#endif
+
+typedef LLMutex LLMutexRootPool;
 
 // Actually a condition/mutex pair (since each condition needs to be associated with a mutex).
-class LL_COMMON_API LLCondition : public LLMutex
+class LLCondition : public LLConditionVariableImpl, public LLMutex
 {
 public:
-	LLCondition(LLAPRPool& parent = LLThread::tldata().mRootPool);
-	~LLCondition();
-	
-	void wait();		// blocks
-	void signal();
-	void broadcast();
-	
-protected:
-	apr_thread_cond_t *mAPRCondp;
+	LLCondition(MUTEX_POOL(native_pool_type& pool = LLThread::tldata().mRootPool)) : 
+		LLMutex(MUTEX_POOL(pool)), 
+		LLConditionVariableImpl(MUTEX_POOL(pool))
+	{}
+	~LLCondition()
+	{}
+	void LL_COMMON_API wait()
+	{
+		if (AIThreadID::in_main_thread_inline())
+			wait_main();
+		else LLConditionVariableImpl::wait(*this);
+	}
+	void signal()		{ LLConditionVariableImpl::notify_one(); }
+	void broadcast()	{ LLConditionVariableImpl::notify_all(); }
+private:
+	void wait_main();	//Cannot be inline. Uses internal fasttimer.
 };
 
-class LL_COMMON_API LLMutexLock
+class LLMutexLock
 {
 public:
-	LLMutexLock(LLMutexBase* mutex)
+	LLMutexLock(LLMutex* mutex)
 	{
 		mMutex = mutex;
-		if(mMutex) mMutex->lock();
+		lock();
+	}
+	LLMutexLock(LLMutex& mutex)
+	{
+		mMutex = &mutex;
+		lock();
 	}
 	~LLMutexLock()
 	{
-		if(mMutex) mMutex->unlock();
+		if (mMutex) mMutex->unlock();
 	}
 private:
-	LLMutexBase* mMutex;
+	LL_COMMON_API void lock();	//Cannot be inline. Uses internal fasttimer.
+	LLMutex* mMutex;
 };
 
-class LL_COMMON_API AIRWLock
+class AIRWLock
 {
 public:
 	AIRWLock(LLAPRPool& parent = LLThread::tldata().mRootPool) :
-		mWriterWaitingMutex(parent), mNoHoldersCondition(parent), mHoldersCount(0), mWriterIsWaiting(false) { }
+		mWriterWaitingMutex(MUTEX_POOL(parent)), mNoHoldersCondition(MUTEX_POOL(parent)), mHoldersCount(0), mWriterIsWaiting(false) { }
 
 private:
 	LLMutex mWriterWaitingMutex;		//!< This mutex is locked while some writer is waiting for access.
@@ -398,7 +567,7 @@ public:
 };
 
 #if LL_DEBUG
-class LL_COMMON_API AINRLock
+class AINRLock
 {
 private:
 	int read_locked;
@@ -409,15 +578,15 @@ private:
 
 	void accessed(void) const
 	{
-	  if (!mAccessed)
-	  {
-		mAccessed = true;
-		mTheadID.reset();
-	  }
-	  else
-	  {
-		llassert_always(mTheadID.equals_current_thread());
-	  }
+		if (!mAccessed)
+		{
+			mAccessed = true;
+			mTheadID.reset();
+		}
+		else
+		{
+			llassert_always(mTheadID.equals_current_thread());
+		}
 	}
 
 public:
@@ -451,22 +620,29 @@ void LLThread::unlockData()
 
 // see llmemory.h for LLPointer<> definition
 
-class LL_COMMON_API LLThreadSafeRefCount
+class LLThreadSafeRefCount
 {
 private:
 	LLThreadSafeRefCount(const LLThreadSafeRefCount&); // not implemented
 	LLThreadSafeRefCount&operator=(const LLThreadSafeRefCount&); // not implemented
 
 protected:
-	virtual ~LLThreadSafeRefCount(); // use unref()
-	
+	virtual ~LLThreadSafeRefCount() // use unref()
+	{
+		if (mRef != 0)
+		{
+			llerrs << "deleting non-zero reference" << llendl;
+		}
+	}
+
 public:
-	LLThreadSafeRefCount();
-	
+	LLThreadSafeRefCount() : mRef(0)
+	{}
+
 	void ref()
 	{
-		mRef++; 
-	} 
+		mRef++;
+	}
 
 	void unref()
 	{
diff --git a/indra/llmath/llcalcparser.h b/indra/llmath/llcalcparser.h
index e3eb2bded..696e4a105 100644
--- a/indra/llmath/llcalcparser.h
+++ b/indra/llmath/llcalcparser.h
@@ -55,44 +55,53 @@ T max_glue(T a, T b)
 	return std::max(a, b);
 }
 
+template <typename T>
 struct lazy_pow_
 {
-	template <typename X, typename Y>
-	struct result { typedef X type; };
- 
-	template <typename X, typename Y>
-	X operator()(X x, Y y) const
+	template<class> struct result;
+	template<class F>
+	struct result
+	{
+		typedef T type;
+	};
+
+	T operator()(T x, T y) const
 	{
 		return std::pow(x, y);
 	}
 };
- 
+
+template <typename T>
 struct lazy_ufunc_
 {
-	template <typename F, typename A1>
-	struct result { typedef A1 type; };
- 
-	template <typename F, typename A1>
-	A1 operator()(F f, A1 a1) const
+	template<class> struct result;
+	template<class F>
+	struct result
 	{
-		return f(a1);
+		typedef T type;
+	};
+
+	T operator()(T(*fn)(T), T x) const
+	{
+		return fn(x);
 	}
 };
- 
+
+template <typename T>
 struct lazy_bfunc_
 {
-	template <typename F, typename A1, typename A2>
-	struct result { typedef A1 type; };
- 
-	template <typename F, typename A1, typename A2>
-	A1 operator()(F f, A1 a1, A2 a2) const
+	template<class> struct result;
+	template<class F>
+	struct result
 	{
-		return f(a1, a2);
+		typedef T type;
+	};
+	double operator()(T(*fn)(T, T), T x, T y) const
+	{
+		return fn(x, y);
 	}
 };
- 
-//} // end namespace anonymous
- 
+
 template <typename FPT, typename Iterator>
 struct grammar
 	: boost::spirit::qi::grammar<
@@ -178,9 +187,9 @@ struct grammar
 		using boost::spirit::qi::no_case;
 		using boost::spirit::qi::_val;
  
-		boost::phoenix::function<lazy_pow_>   lazy_pow;
-		boost::phoenix::function<lazy_ufunc_> lazy_ufunc;
-		boost::phoenix::function<lazy_bfunc_> lazy_bfunc;
+		boost::phoenix::function< lazy_pow_<FPT> >   lazy_pow;
+		boost::phoenix::function< lazy_ufunc_<FPT> > lazy_ufunc;
+		boost::phoenix::function< lazy_bfunc_<FPT> > lazy_bfunc;
  
 		expression =
 			term                   [_val =  _1]
diff --git a/indra/llmessage/aicurl.cpp b/indra/llmessage/aicurl.cpp
index 99683e0c5..68328c394 100644
--- a/indra/llmessage/aicurl.cpp
+++ b/indra/llmessage/aicurl.cpp
@@ -1300,7 +1300,7 @@ static int const HTTP_REDIRECTS_DEFAULT = 16;	// Singu note: I've seen up to 10
 												// This limit is only here to avoid a redirect loop (infinite redirections).
 
 LLChannelDescriptors const BufferedCurlEasyRequest::sChannels;
-LLMutex BufferedCurlEasyRequest::sResponderCallbackMutex;
+LLGlobalMutex BufferedCurlEasyRequest::sResponderCallbackMutex;
 bool BufferedCurlEasyRequest::sShuttingDown = false;
 AIAverage BufferedCurlEasyRequest::sHTTPBandwidth(25);
 
diff --git a/indra/llmessage/aicurlprivate.h b/indra/llmessage/aicurlprivate.h
index be3d90e7d..9af27a567 100644
--- a/indra/llmessage/aicurlprivate.h
+++ b/indra/llmessage/aicurlprivate.h
@@ -430,7 +430,7 @@ class BufferedCurlEasyRequest : public CurlEasyRequest {
 
   public:
 	static LLChannelDescriptors const sChannels;		// Channel object for mInput (channel out()) and mOutput (channel in()).
-	static LLMutex sResponderCallbackMutex;				// Locked while calling back any overridden ResponderBase::finished and/or accessing sShuttingDown.
+	static LLGlobalMutex sResponderCallbackMutex;		// Locked while calling back any overridden ResponderBase::finished and/or accessing sShuttingDown.
 	static bool sShuttingDown;							// If true, no additional calls to ResponderBase::finished will be made anymore.
 	static AIAverage sHTTPBandwidth;					// HTTP bandwidth usage of all services combined.
 
diff --git a/indra/llmessage/aicurlthread.cpp b/indra/llmessage/aicurlthread.cpp
index 19f190b6b..b7057fe11 100644
--- a/indra/llmessage/aicurlthread.cpp
+++ b/indra/llmessage/aicurlthread.cpp
@@ -1098,10 +1098,10 @@ void AICurlThread::wakeup_thread(bool stop_thread)
   if (stop_thread)
 	mRunning = false;			// Thread-safe because all other threads were already stopped.
 
-  // Note, we do not want this function to be blocking the calling thread; therefore we only use tryLock()s.
+  // Note, we do not want this function to be blocking the calling thread; therefore we only use try_lock()s.
 
   // Stop two threads running the following code concurrently.
-  if (!mWakeUpMutex.tryLock())
+  if (!mWakeUpMutex.try_lock())
   {
 	// If we failed to obtain mWakeUpMutex then another thread is (or was) in AICurlThread::wakeup_thread,
 	// or curl was holding the lock for a micro second at the start of process_commands.
@@ -1115,7 +1115,7 @@ void AICurlThread::wakeup_thread(bool stop_thread)
   }
 
   // Try if curl thread is still awake and if so, pass the new commands directly.
-  if (mWakeUpFlagMutex.tryLock())
+  if (mWakeUpFlagMutex.try_lock())
   {
 	mWakeUpFlag = true;
 	mWakeUpFlagMutex.unlock();
diff --git a/indra/llmessage/llpumpio.cpp b/indra/llmessage/llpumpio.cpp
index 588265f97..a2b63e8e1 100644
--- a/indra/llmessage/llpumpio.cpp
+++ b/indra/llmessage/llpumpio.cpp
@@ -56,7 +56,7 @@
 
 // constants for poll timeout. if we are threading, we want to have a
 // longer poll timeout.
-#if LL_THREADS_APR
+#if LL_THREADS_PUMPIO
 static const S32 DEFAULT_POLL_TIMEOUT = 1000;
 #else
 static const S32 DEFAULT_POLL_TIMEOUT = 0;
@@ -168,24 +168,21 @@ LLPumpIO::LLPumpIO(void) :
 	mPollset(NULL),
 	mPollsetClientID(0),
 	mNextLock(0),
+#if LL_THREADS_PUMPIO
+	mPool(),
+	LLMutex mChainsMutex(initPool()),
+	LLMutex mCallbackMutex(initPool()),
+#endif
 	mCurrentPoolReallocCount(0),
-	mChainsMutex(NULL),
-	mCallbackMutex(NULL),
 	mCurrentChain(mRunningChains.end())
 {
-	mCurrentChain = mRunningChains.end();
-
-	initialize();
+#if !LL_THREADS_PUMPIO
+	initPool();
+#endif
 }
 
 LLPumpIO::~LLPumpIO()
 {
-#if LL_THREADS_APR
-	if (mChainsMutex) apr_thread_mutex_destroy(mChainsMutex);
-	if (mCallbackMutex) apr_thread_mutex_destroy(mCallbackMutex);
-#endif
-	mChainsMutex = NULL;
-	mCallbackMutex = NULL;
 	if(mPollset)
 	{
 //		lldebugs << "cleaning up pollset" << llendl;
@@ -218,8 +215,8 @@ bool LLPumpIO::addChain(chain_t const& chain, F32 timeout)
 		info.mChainLinks.push_back(link);
 	}
 
-#if LL_THREADS_APR
-	LLScopedLock lock(mChainsMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mChainsMutex);
 #endif
 	mPendingChains.push_back(info);
 	return true;
@@ -258,8 +255,8 @@ bool LLPumpIO::addChain(
 			break;
 		}
 	}
-#if LL_THREADS_APR
-	LLScopedLock lock(mChainsMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mChainsMutex);
 #endif
 	mPendingChains.push_back(info);
 	return true;
@@ -403,8 +400,8 @@ void LLPumpIO::clearLock(S32 key)
 	// therefore won't be treading into deleted memory. I think we can
 	// also clear the lock on the chain safely since the pump only
 	// reads that value.
-#if LL_THREADS_APR
-	LLScopedLock lock(mChainsMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mChainsMutex);
 #endif
 	mClearLocks.insert(key);
 }
@@ -469,8 +466,8 @@ void LLPumpIO::pump(const S32& poll_timeout)
 	PUMP_DEBUG;
 	if(true)
 	{
-#if LL_THREADS_APR
-		LLScopedLock lock(mChainsMutex);
+#if LL_THREADS_PUMPIO
+		LLMutexLock lock(mChainsMutex);
 #endif
 		// bail if this pump is paused.
 		if(PAUSING == mState)
@@ -738,8 +735,8 @@ void LLPumpIO::pump(const S32& poll_timeout)
 
 //bool LLPumpIO::respond(const chain_t& pipes)
 //{
-//#if LL_THREADS_APR
-//	LLScopedLock lock(mCallbackMutex);
+//#if LL_THREADS_PUMPIO
+//	LLMutexLock lock(mCallbackMutex);
 //#endif
 //	LLChainInfo info;
 //	links_t links;
@@ -752,8 +749,8 @@ bool LLPumpIO::respond(LLIOPipe* pipe)
 {
 	if(NULL == pipe) return false;
 
-#if LL_THREADS_APR
-	LLScopedLock lock(mCallbackMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mCallbackMutex);
 #endif
 	LLChainInfo info;
 	LLLinkInfo link;
@@ -773,8 +770,8 @@ bool LLPumpIO::respond(
 	if(!data) return false;
 	if(links.empty()) return false;
 
-#if LL_THREADS_APR
-	LLScopedLock lock(mCallbackMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mCallbackMutex);
 #endif
 
 	// Add the callback response
@@ -793,8 +790,8 @@ void LLPumpIO::callback()
 	//llinfos << "LLPumpIO::callback()" << llendl;
 	if(true)
 	{
-#if LL_THREADS_APR
-		LLScopedLock lock(mCallbackMutex);
+#if LL_THREADS_PUMPIO
+		LLMutexLock lock(mCallbackMutex);
 #endif
 		std::copy(
 			mPendingCallbacks.begin(),
@@ -821,8 +818,8 @@ void LLPumpIO::callback()
 
 void LLPumpIO::control(LLPumpIO::EControl op)
 {
-#if LL_THREADS_APR
-	LLScopedLock lock(mChainsMutex);
+#if LL_THREADS_PUMPIO
+	LLMutexLock lock(mChainsMutex);
 #endif
 	switch(op)
 	{
@@ -838,14 +835,11 @@ void LLPumpIO::control(LLPumpIO::EControl op)
 	}
 }
 
-void LLPumpIO::initialize(void)
+LLAPRPool& LLPumpIO::initPool()
 {
-	mPool.create();
-#if LL_THREADS_APR
-	// SJB: Windows defaults to NESTED and OSX defaults to UNNESTED, so use UNNESTED explicitly.
-	apr_thread_mutex_create(&mChainsMutex, APR_THREAD_MUTEX_UNNESTED, mPool());
-	apr_thread_mutex_create(&mCallbackMutex, APR_THREAD_MUTEX_UNNESTED, mPool());
-#endif
+	if (!mPool)
+		mPool.create();
+	return mPool;
 }
 
 void LLPumpIO::rebuildPollset()
diff --git a/indra/llmessage/llpumpio.h b/indra/llmessage/llpumpio.h
index 0d1387257..309792c80 100644
--- a/indra/llmessage/llpumpio.h
+++ b/indra/llmessage/llpumpio.h
@@ -42,7 +42,7 @@
 #include "llrun.h"
 
 // Define this to enable use with the APR thread library.
-//#define LL_THREADS_APR 1
+//#define LL_THREADS_PUMPIO 1
 
 // some simple constants to help with timeouts
 extern const F32 DEFAULT_CHAIN_EXPIRY_SECS;
@@ -382,16 +382,13 @@ protected:
 	LLAPRPool mCurrentPool;
 	S32 mCurrentPoolReallocCount;
 
-#if LL_THREADS_APR
-	apr_thread_mutex_t* mChainsMutex;
-	apr_thread_mutex_t* mCallbackMutex;
-#else
-	int* mChainsMutex;
-	int* mCallbackMutex;
+#if LL_THREADS_PUMPIO
+	LLMutex mChainsMutex;
+	LLMutex mCallbackMutex;
 #endif
 
 protected:
-	void initialize();
+	LLAPRPool& initPool();
 
 	current_chain_t removeRunningChain(current_chain_t& chain) ;
 	/**